blockdag // AI inference // optimistic proofs // proof-of-model

KERYX

Free Intelligence Engine

The first BlockDAG built for decentralized AI inference. Optimistic proofs. Unstoppable intelligence.

keryx-miner v0.3.3
$ keryx-miner --network mainnet --gpu 0 --very-high
# verifying model integrity via IPFS CID...
Mistral-7B-v0.3 [sha256:8c2fea...] — 1.0 KRX/req
GLM-4-9B [sha256:fa2f13...] — 1.5 KRX/req
Qwen3.6-27B [sha256:b8bdc0...] — 2.5 KRX/req
Kimi-Linear-48B [sha256:3dc093...] — 4.0 KRX/req
GHOSTDAG sync — DAA 15,551,204 — 10 BPS
inference complete — uploading result to IPFS...
$ AiResponse submitted — CID: QmX7a2f... — +4.0 KRX

The Centralized Black Box

Three corporations control the filters of global AI. On-chain agents can't run local models. They need a decentralized inference layer.

// threat

Silent Censorship

AI providers quietly modify model outputs to align with corporate policies. No transparency, no audit trail, no recourse.

// threat

Prompt Surveillance

Every query passes through centralized servers. Your questions, strategies, and private data — logged, analyzed, monetized.

// keryx

Sovereign AI

The model runs across thousands of sovereign miners. No kill switch. Intelligence becomes a public infrastructure.

// keryx

Optimistic Verifiability

Miners lock tokens on every response. Anyone can challenge. Cheaters lose everything. Honesty is the only profitable strategy.

Optimistic Proof of Inference

Respond first, verify only when challenged. Fast, practical, cryptographically secure.

01

Request

User or AI agent submits an AI_Request to the mempool with model ID and fees.

02

Inference

Miner runs the SLM on their GPU, publishes the response instantly, and locks a collateral as guarantee.

03

Challenge Window

For 36,000 blocks (~1h), any verifier can re-run the deterministic model and dispute the result. Fraud proven on-chain via re-execution — miner's escrow is slashed.

04

Finality

No challenge — result is final, miner earns reward. Fraud proven — miner is slashed, challenger is rewarded.

> on-chain inference lifecycle
AiRequest tx
fee locked in escrow
AiResponse tx
miner commits IPFS CID
challenge window
36,000 blocks (~1h)
fee released
to miner wallet

> cold_start_bootstrap

No users yet? The protocol self-generates synthetic inference tasks derived from the previous block hash. Miners stay active, the network stays warm, infrastructure is battle-tested before the first real request.

The Weights Are the Work

Keryx's proof-of-work isn't arbitrary hashing — it's the model itself. To win a block, a miner's GPU must actually hold a full AI model in memory.

01

Possession

Mining requires a model's weights resident in VRAM. No idle hashing on a bare GPU — the work is the model.

02

The Weight-Walk

The proof-of-work walks the resident weights — each read's location derived from the last. A partial copy can't precompute or stream fast enough: miss a fraction of the weights and you slow down by orders of magnitude.

03

One GPU = One Model

The largest model your VRAM holds sets your tier — from EXAONE-4.0-1.2B on a 2 GB card to Kimi-Linear-48B on a 32 GB card. Heavier tier, larger block reward.

04

Trustless

Verification is byte-exact and fully deterministic: a per-tier weight Merkle root plus Fiat-Shamir chunk openings, checked from the block header alone. Pool-safe and fork-proof — no trust in the miner.

> possession, not correctness

PoM proves a miner holds the model — a hardware-enforced guarantee at the consensus level. Whether a given answer was computed honestly is the job of the optimistic OPoI challenge layer above. Two complementary guarantees: PoM secures the who, OPoI secures the what.

Verified AI Models

A fully uncensored lineup — no corporate filters, no refusals. Every model is identified by sha2-256(weight_file) — the hash extracted directly from its IPFS CID. A miner cannot fake which model it runs. Requesters set a priority_fee above the minimum to be served first.

// 1.2B params · uncensored

EXAONE-4.0 1.2B

0.5 KRXmin / request
sha256: 300a99b3...c4b810

Ultra-light and unfiltered. Runs on a 2 GB GPU — the entry tier for high-frequency agent calls and synthetic workloads.

// 7B params · uncensored

Mistral-7B v0.3

1.0 KRXmin / request
sha256: 8c2fea60...581e0a

Fast, lightweight, unfiltered. Ideal for short classification tasks, synthetic workloads, and high-frequency agent calls.

// 9B params · uncensored

GLM-4 9B

1.5 KRXmin / request
sha256: fa2f13be...f2140a

Balanced reasoning and speed, no corporate filters. Suitable for complex queries, on-chain agent decisions, and multi-step logic.

// 27B params · uncensored

Qwen3.6 27B

2.5 KRXmin / request
sha256: b8bdc01f...284746

High-fidelity reasoning at scale, unrestricted. A hybrid SSM/attention architecture for deep analysis, code generation, and long-context understanding.

// 48B params (MoE) · uncensored

Kimi-Linear 48B

4.0 KRXmin / request
sha256: 3dc09358...dffddb

State-of-the-art open MoE, fully unfiltered — 48B total, 3B active per token. The Q4_K_M quant fits a single 32 GB card. Reserved for the most demanding autonomous agent workflows.

> gpu_tiers — pick your flag
--very-lightEXAONE-4.0-1.2B2 GB+ VRAM
--lightMistral-7B-v0.38 GB+ VRAM
(default)GLM-4-9B12 GB+ VRAM
--highQwen3.6-27B24 GB+ VRAM
--very-highKimi-Linear-48B32 GB+ VRAM

> priority_fee

Requesters can bid above the minimum to jump the queue. Miners serve the highest-paying requests first — creating a real market for inference priority. Every fee is paid in $KRX, creating direct, measurable buy pressure from actual AI usage.

Who Needs Keryx?

Whether for humans seeking truth without filters or autonomous agents operating on-chain, Keryx is the trustless backbone of decentralized intelligence.

>

Autonomous On-Chain Agents

Solana and Ethereum AI agents are, today, fully centralized via Web2 APIs. Smart contracts are deterministic — they cannot reason, interpret language, or make decisions, so the moment an agent has to think it calls OpenAI or Anthropic and inherits their kill switch. With Keryx, an agent broadcasts an AiRequest on-chain, a miner executes the inference, and the result comes back as a verifiable AiResponse — no HTTP call, no API key, no corporate gateway. The response is backed by real economic stake: fraudulent answers are slashable. The chain is the oracle.

>

Uncensored Knowledge

Legal, medical, or political queries without corporate filters. A neutral AI that answers factually regardless of the topic.

>

Encrypted Prompt Submission

Roadmap: users encrypt their prompt with the miner's public key before broadcasting. The miner decrypts and runs inference locally — the plaintext prompt never appears on-chain in any form. The OPoI commitment covers the result, preserving full fraud-proof guarantees on the output while keeping the query private.

Tech Stack

rusty-kaspa fork
GHOSTDAG BlockDAG at 10 BPS. Memory-hard PoW to keep mining GPU-only.
inference engine
In-process GGUF inference engine. Runs 5 verified models (1.2B → 48B) natively on GPU. Model identity = sha2-256(weight_file), derived from the IPFS CID — cryptographically unforgeable.
fraud proofs
Deterministic re-execution: the node re-runs the fixed-point model to verify the miner's commitment. OPoI commitment = response_ipfs_cid[2..34], the SHA2-256 of the result content. Same security model as Ethereum L2 fraud proofs.
IPFS
Model weights and inference results stored off-chain. Responses committed as 34-byte CIDv0 on-chain — unlimited length, content-addressed, decentralized. Miners run kubo automatically.
collateral layer
Miners lock KRX as guarantee. Dishonest inference = slashed collateral.

$KRX Economics

Designed for long-term scarcity. Every transaction destroys value. Every block splits reward between miners, infrastructure, and protocol burn.

fee burn
100% of transaction fees are permanently destroyed — fees reduce circulating supply on every block.
min fee
0.3 KRX minimum per transaction. Spam-resistant and economically sound — every on-chain action has a real deflationary cost.
Block Reward — 5.4 KRX
75%
14.4%
5%
5.6%
4.05 KRX — Miner
Sent immediately to the miner's wallet on block confirmation
0.78 KRX — Escrow
Locked as collateral, recovered after the 1h challenge window closes
0.27 KRX — R&D
Allocated to protocol research, model maintenance, and development
0.30 KRX — Burned
Escrow transaction fees — permanently removed from supply

Roadmap

phase_1Genesis — Completed
  • Full Kaspa → Keryx rebrand (binary, ports, address prefix, genesis block)
  • KeryxHash (KERYX_MATRIX_SALT + wave_mix)
  • OPoI tag in every coinbase (format validation)
  • TinyLlama-1.1B inference in miner (candle-transformers)
  • Explorer + API (real-time, PostgreSQL indexed)
  • Richlist, top miners on explorer
  • Web wallet integration
phase_2Economy — Completed
  • 20% escrow routing (CSV-locked, auto-claim in miner)
  • AiRequest / AiResponse / AiChallenge transactions
  • On-chain fraud proof (fixed-point re-execution)
  • Automated challenger daemon (keryx-challenger)
  • P2P attack mitigations (autoban)
  • Escrow hard-lock: prevent withdrawal before challenge window closes
  • Transaction fee burn (100% of fees destroyed)
  • Minimum transaction fee: 0.3 KRX
phase_3Multi-Model Oracle — Deployed (DAA 15,550,000)
  • Multi-model support: TinyLlama-1.1B, DeepSeek-R1-8B, DeepSeek-R1-32B, LLaMA-3.3-70B
  • Per-model inference_reward minimums enforced at consensus level
  • priority_fee separate from inference_reward (burned vs miner-paid)
  • Token-count surcharge (0.05 KRX per 64-token increment)
  • Model capability announcement in coinbase extra_data (/ai:cap:)
  • Consensus enforcement: AiResponse rejected if model not declared in caps
  • IPFS result storage: AiResponse stores CIDv0 instead of inline text
  • model_id = SHA-256(weight file) = IPFS CIDv0 content hash
  • IPFS for model weight distribution (miner downloads from Keryx IPFS gateway)
  • CUDA inference support (candle-core, VRAM-aware multi-model placement)
phase_4Proof-of-Model & Holder Economics — Deployed (DAA 37,780,000)
  • PoM: KeryxHash matrix-multiply replaced by a data-dependent walk over the resident model weights (possession proof, one GPU = one tier)
  • Byte-exact deterministic verification via per-tier weight Merkle root R_T + Fiat-Shamir chunk openings (pool-safe, fork-proof)
  • Zero-copy: PoM walk shares the inference weights resident in VRAM (no second copy)
  • Uncensored five-tier lineup: Qwen3-1.7B, Gemma-3-4B, Dolphin-3.0-8B, Qwen3-32B, LLaMA-3.3-70B (Q2) — one GPU, one tier
  • Tier-reward: miner cut scaled by proven PoM tier (68/76/84/92/100%), delta burned
  • Holder-reward: miner cut scaled by KRX held relative to windowed production, delta burned
  • CUDA mining engine (PoM weight-walk + inference on a custom CUDA kernel)
phase_5Scaling & Sovereign Intelligence — Research
  • Multi-GPU VRAM pooling: aggregate a whole rig (e.g. 6× 8 GB cards = 48 GB) via layer-split inference, so mining rigs serve models no single card can hold
  • Private end-to-end encrypted inference: prompts sealed to the serving miner's key; the public feed shows only metadata (model, fee, 🔒)
  • Stateful models: retrieval-augmented memory (tractable) and tool/network access (private track only — non-reproducible calls cannot be fraud-proven)
  • Continual fine-tuning: tractable per-miner, but a shared evolving model is an open research problem (fine-tuning changes model_id; agreeing on new weights needs deterministic federated training)
phase_6The Cross-Chain Oracle
  • Ethereum bridge: smart contract → AiRequest → Keryx → result → Ethereum callback
  • Solana bridge (same pattern)
  • On-chain agent demo: autonomous Ethereum agent calling Keryx for decisions
  • Encrypted prompt submission (miner pubkey encryption)
  • Cross-block AiResponse enforcement (cap enforcement for requests in older blocks)

Intelligence belongs to no one.

"Intelligence is the message. Keryx is the messenger."
// fork. mine. liberate. //